Model Context Protocol

Bilingual Mix, callable by your AI agent.

One remote MCP endpoint. Ask Claude, ChatGPT, Cursor or any MCP client to turn a page of parallel text into a sentence-by-sentence bilingual track — the same voices and the same quota you get on this site.

Endpoint

Streamable HTTP, stateless. No key required to try it — the free tier is per-IP.

https://bilingualmix.com/mcp

Connect in one command

Claude Code:

claude mcp add --transport http bilingual-mix https://bilingualmix.com/mcp

Claude Desktop / Cursor / VS Code — add to the MCP config:

{
  "mcpServers": {
    "bilingual-mix": {
      "type": "http",
      "url": "https://bilingualmix.com/mcp"
    }
  }
}

Some clients only accept local (stdio) servers. In that case use an HTTP-to-stdio bridge such as mcp-remote. The endpoint itself takes no configuration on our side.

What the agent can do

Tool	What it does	Quota
`create_bilingual_audio`	Turns sentence pairs into one alternating A→B track. Returns a 6-hour WAV link plus a per-sentence timeline in seconds.	2 runs/day
`align_audio`	Takes two public audio URLs, transcribes both and aligns them into sentence pairs with timestamps.	2 runs/day
`translate_lines`	Fills in the missing side when you only have Chinese or only have English.	Members
`get_quota`	Sign-in state, minutes used and left, per-run limits.	Free
`list_plans`	Pricing tiers and allowances.	Free

How the free trial and sign-in work

1 · Try itNo account: 2 runs and 20 minutes a day, up to 8 sentence pairs per run.

2 · Sign inWhen the free quota is spent, the server replies 401 and the client opens the OAuth consent page. Use the account you already have here.

3 · Keep goingFree accounts get 4 runs / 40 min a day. Members get 500 minutes a month and up to 45 minutes per run.

Authorization is standard OAuth 2.1 with PKCE (S256) — no API key to copy, and no password ever reaches the agent. Access tokens last one hour; you can revoke by signing out of your account.

Example prompt

# once connected, just ask:
Use bilingual-mix. Here are 12 Chinese lines and their English
translations. First run it with dry_run to show me the cost,
then generate the audio and give me the download link
and the per-sentence timeline.

dry_run never spends quota, so the agent can check length and price before generating.

Or use the web studio See plans

Discover metadata: /.well-known/oauth-protected-resource · /.well-known/oauth-authorization-server. Machine-readable summary in llms.txt.